Kubiya

What is Kubiya?

Kubiya is an AI agent orchestration platform built for DevOps, platform engineering, and infrastructure teams that want automation with guardrails. It is not trying to be a general chatbot for the whole company. The product is designed around operational work, things like provisioning infrastructure, managing Kubernetes, triggering CI/CD workflows, responding to incidents, and coordinating approvals inside tools teams already use, especially Slack and Microsoft Teams.

We found Kubiya’s story starts with a familiar enterprise problem. Teams want AI to reduce repetitive engineering work, but most AI tools behave like stateless assistants. They answer a question, then forget everything. They also tend to sit outside the systems where real work happens. Kubiya was built to solve that gap by acting as a control layer across cloud infrastructure, observability, CI/CD, and internal workflows, while keeping deterministic execution, audit logs, RBAC, and policy enforcement front and center. The company was founded in 2022 by Amit Eyal Govrin and Shaked Askayo, drawing on backgrounds from AWS and fintech, and has raised $12 million in seed funding. It is headquartered in San Jose, with additional presence in Tel Aviv.

The customer list tells you who this is really for. Kubiya names Project44, Morse, A+E Networks, Verana Health, Ford, Atlassian, and Databricks among its users. That mix matters. It suggests the platform is aimed at mid-size and large organizations where infrastructure work is complex, compliance matters, and AI has to fit into existing governance instead of bypassing it.

Key Features

• AI agent orchestration for DevOps: Kubiya coordinates agents, workers, queues, and workflows so teams can run operational tasks across infrastructure systems instead of just chatting with a model. This matters because the hard part in enterprise AI is rarely text generation, it is getting repeatable work done safely across many tools and environments.

• Deterministic execution with task tracking: Work is tracked through task states like Pending, Running, Waiting for Input, Completed, and Failed using Kubiya’s Task Kanban model. That gives teams a way to inspect and measure AI work like they already measure Jira work, instead of relying on vague claims about productivity.

• Cognitive Memory: Kubiya gives agents persistent memory across sessions, with shared context inside the same environment and isolation between environments like staging and production. Under the hood, the system uses PostgreSQL, pgvector, Neo4j, and semantic search, which matters because teams stop re-explaining the same context every time they ask for help.

• Multi-runtime model support: The Agno runtime supports 100+ LLM providers through LiteLLM, while Claude Code is optimized for code-heavy tasks on Anthropic models. In practice, this gives teams a choice between model flexibility and code-specialized behavior without rebuilding their agent layer every time model preferences change.

• Deep DevOps integrations: Kubiya integrates with Kubernetes, Terraform, GitHub, Jenkins, CircleCI, Prometheus, Datadog, PagerDuty, Slack, Microsoft Teams, and more. The value is in the depth, not just the logos, because workflows can cross from a natural-language request to provisioning, monitoring, alerting, and ticketing in one chain.

• Role-based access control and just-in-time approvals: Kubiya enforces RBAC at the action, tool, resource, and environment level, and supports human approvals for sensitive actions. That matters in production environments where teams want AI help, but not silent database changes or unreviewed deployments.

• Policy-as-code with OPA: Organizations can define rules for who can run what, when tasks can execute, and what validations have to happen first. This gives platform and security teams a way to govern AI behavior with the same discipline they apply to infrastructure and deployment policy.

• Self-hosted and air-gapped deployment options: Kubiya can run cloud-hosted, self-hosted, or in air-gapped environments. For regulated industries or companies with strict data residency rules, this is often the difference between “interesting demo” and “possible to adopt.”

• Custom tools and MCP support: Teams can build their own tools in isolated Docker containers and expose systems through the Model Context Protocol. This matters for enterprises with internal systems that no vendor supports out of the box.

• Infrastructure-as-code and API access: Kubiya offers a REST control plane API, Python SDKs, CLI access, and an official Terraform provider. That gives platform teams a path to manage Kubiya the same way they manage the rest of their stack, with version control and reproducible configuration.

Use Cases

Kubiya is strongest when a company wants AI to operate inside engineering workflows, not just advise on them. One recurring example in the research is Kubernetes operations. Teams use Kubiya to scale resources during high-traffic events, monitor cluster health, and notify stakeholders in Slack, all through predefined workflows rather than improvised AI-generated commands. That distinction matters. In Kubernetes, creativity is usually less valuable than consistency.

CI/CD automation is another major thread. Kubiya lets engineers start with natural language, such as asking for a staging environment or triggering a deployment flow, then turns that into a multi-step workflow spanning infrastructure, CI/CD, and communication tools. The practical benefit is not just speed. It lowers the amount of platform-specific knowledge an engineer needs to get routine delivery work done.

Incident response is where the platform’s governance story becomes more concrete. The research describes Kubiya detecting patterns in failed builds or outages, interpreting logs, and triggering recovery workflows like restarting deployments, rolling back changes, or muting noisy alerts. Because those actions are tied to RBAC, approvals, and audit trails, the platform is trying to reduce on-call burden without creating a new class of opaque automation risk.

The customer list gives some confidence that this is not a narrow startup-only product. Project44, Ford, Atlassian, Databricks, Verana Health, Morse, and A+E Networks all suggest different operational environments and compliance expectations. The research does not give detailed customer-by-customer rollout metrics, but it does point to a broad pattern: teams use Kubiya to remove repetitive operational work while keeping human judgment in the loop.

There is also a cost optimization angle. Kubiya’s AIOps use cases include matching workloads to cheaper infrastructure options, combining billing data with utilization, and forecasting future cloud spend. The research cites Gartner benchmarks of 20 to 30 percent reductions in operational and cloud costs for enterprises using this style of AI-driven infrastructure optimization. That is a benchmark, not a Kubiya-specific case study, but it helps frame the kind of ROI buyers are looking for.

Strengths and Weaknesses

Strengths:

Kubiya looks unusually serious about governance for an AI product. A lot of agent tools are great at demos and weak on approval flows, auditability, and policy enforcement. Kubiya was built around those concerns from the start, with OPA policies, RBAC, just-in-time approvals, and immutable audit logs. For a platform team in a regulated or high-risk environment, that is a meaningful difference from frameworks like LangChain or CrewAI, which offer flexibility but leave much more of the operational hardening to the customer.

It also is known for meeting engineers where they already work. The Slack and Teams integration is not just a convenience feature. It changes adoption patterns because people can request actions, approve tasks, and receive updates without switching into another interface. Combined with integrations for GitHub, Jenkins, Terraform, Kubernetes, Datadog, and PagerDuty, Kubiya feels closer to an orchestration layer than a standalone AI app.

The memory model is another real differentiator. Most AI tools still behave like they are starting from zero every time. Kubiya’s Cognitive Memory gives agents a shared memory inside an environment, which means they can recall past context, solutions, and relationships. For teams dealing with recurring incidents, repeated deployment patterns, or long-running operational processes, that can reduce a lot of repetitive prompting.

Deployment flexibility is a practical strength. Cloud-hosted is there for speed, but self-hosted and air-gapped options broaden the set of companies that can consider the platform. That matters more than flashy model support in industries where data handling rules decide the shortlist before features do.

Weaknesses:

Kubiya is not the simplest product in this category, because it is solving a harder problem. The architecture includes control planes, workers, queues, environments, teams, policies, runtimes, and custom tools. For mature platform teams, that structure is the point. For smaller teams looking for quick automation without much setup discipline, it may feel heavy compared with simpler workflow tools or even direct use of cloud-native scripts.

Pricing is also less transparent than many buyers will want. Kubiya uses yearly Agentic Engineering Hours retainers instead of public monthly plans. That can fit enterprise procurement, but it makes early budgeting harder and introduces a use-it-or-lose-it dynamic since unused hours expire unless rollover terms are negotiated. Teams comparing Kubiya with open-source frameworks may find the managed convenience appealing, but they will also notice the trade-off in flexibility and cost visibility.

There is a broader trade-off between platform completeness and customization. Compared with LangChain, CrewAI, or LangGraph, Kubiya gives up some low-level freedom in exchange for managed infrastructure, security controls, and ready-made integrations. If your team wants to invent highly custom agent architectures and owns the operational burden, a framework may still be the better fit.

Finally, some of the strongest value claims in the research are directional rather than deeply quantified. We found named enterprise customers and strong architectural detail, but fewer public customer stories with exact before-and-after metrics tied specifically to Kubiya deployments. Buyers who need hard ROI proof will probably want a pilot.

Pricing

• Pilot program: Custom, 2-month pilot Kubiya offers a 2-month pilot for teams that want to validate the product before a longer commitment. That fits the reality of enterprise AI buying, where security review and operational proof usually come before budget approval.

• Annual AEH retainer: Custom pricing Kubiya sells through Agentic Engineering Hours, a yearly retainer model where customers consume agent work over time. Platform access and support are included, so the pricing is tied to usage rather than feature gating.

• Enterprise: Custom Larger customers can negotiate custom terms, including committed usage, pricing tiers, support, and rollover treatment for unused hours. Self-hosted and hybrid deployment requirements likely push most serious buyers into this tier.

The important pricing story is what is not public. We did not find list pricing, so most buyers will need to talk to sales. The unusual part is the AEH model itself. It is closer to buying a pool of operational capacity than buying seats. That may work well for companies with fluctuating automation demand, but teams should ask carefully about expiration, overages, and what actually counts as an engineering hour. Compared with open-source frameworks, Kubiya will cost more in direct spend. Compared with building and governing the same system internally, it may cost less in staffing and risk.

Alternatives

LangChain LangChain is for teams that want a developer framework, not a managed enterprise control plane. It gives engineers a lot of freedom to build custom LLM workflows and agent behavior, but security, hosting, governance, and operational reliability are largely the customer’s job. A team might choose LangChain over Kubiya if it has strong in-house AI engineering talent and wants maximum control. It might choose Kubiya if the real need is production governance and integration with DevOps systems.

CrewAI CrewAI is often considered when the goal is multi-agent collaboration and role-based agent design. It is appealing to builders who want to orchestrate specialist agents and own the logic themselves. Compared with Kubiya, it is more of a toolkit and less of a finished enterprise platform. If your main challenge is experimenting with agent collaboration patterns, CrewAI is attractive. If your main challenge is getting that work approved by security and used by platform teams, Kubiya has a clearer story.

LangGraph LangGraph is a good fit for teams that want graph-based control over agent workflows and state transitions. It is more explicit and programmable than many agent abstractions, which makes it useful for complex custom applications. But again, it is a framework choice. Kubiya sits higher in the stack, with deployment models, governance, integrations, and enterprise controls already packaged.

Jira, Asana, ClickUp, monday.com These tools are not direct substitutes for Kubiya, but they often appear in adjacent evaluations because teams use them to coordinate operational work. They are much better at managing human tasks than running governed AI agents across infrastructure. If your bottleneck is planning and project visibility, these tools may be enough. If your bottleneck is actually executing infrastructure and DevOps work through AI with approvals and audit trails, Kubiya is solving a different problem.

Internal scripts and platform tooling For some companies, the real alternative is not another vendor. It is Bash scripts, Terraform modules, GitHub Actions, Jenkins jobs, and internal platform portals glued together over years. That stack can be cheaper on paper and highly tailored, but it often becomes brittle and hard to govern as use cases multiply. Kubiya’s pitch is that it becomes the orchestration layer above those systems rather than replacing them outright.

FAQ

What is Kubiya used for?

Kubiya is used to automate DevOps and infrastructure work, things like Kubernetes operations, CI/CD workflows, incident response, cloud provisioning, and internal engineering tasks. It is built for teams that need AI to act inside real systems, not just answer questions.

Who is Kubiya built for?

It is aimed at platform engineering, DevOps, SRE, and security-conscious enterprise teams. Non-technical users can also interact with it through Slack or Teams, but the core value is operational automation.

How do I get started?

Most teams start with a pilot. From there, they connect Kubiya to tools like Slack, GitHub, Kubernetes, Terraform, and observability systems, then define environments, permissions, and approval policies.

How long does it take to set up?

The research does not give a standard setup timeline. In practice, cloud-hosted deployments should move faster, while self-hosted or air-gapped setups will take longer because they involve infrastructure, security review, and policy configuration.

Does Kubiya work with Slack?

Yes. Slack is one of its key interfaces, used for requesting actions, receiving updates, and routing approvals. Microsoft Teams is also supported.

Can Kubiya run in our own environment?

Yes. Kubiya supports cloud-hosted, self-hosted, and air-gapped deployments. That is especially relevant for regulated industries or companies with strict security boundaries.

What AI models does Kubiya support?

Through the Agno runtime and LiteLLM, Kubiya supports more than 100 model providers, including OpenAI, Anthropic, Google, and Azure. It also offers a Claude Code runtime for code-focused tasks on Anthropic models.

What makes Kubiya different from a chatbot?

A chatbot mostly answers prompts. Kubiya tracks tasks, orchestrates tools, applies policy, stores memory, and executes work across infrastructure systems with approvals and audit logs.

Does Kubiya have memory across sessions?

Yes. Its Cognitive Memory system lets agents remember context and share knowledge within the same environment. Production and staging memories stay isolated from each other.

Is Kubiya secure enough for enterprise use?

It was clearly designed with that goal. The platform includes RBAC, OPA-based policy enforcement, just-in-time approvals, immutable audit logs, secrets management, and support for self-hosted deployments. The research also cites SOC 2 Type II, GDPR, CCPA, HIPAA-ready configurations, and ISO 27001.

How is Kubiya priced?

Kubiya uses a yearly Agentic Engineering Hours retainer model rather than public per-seat or per-month pricing. There is also a 2-month pilot option, and enterprise terms are negotiated directly.

What are the main trade-offs?

Kubiya gives teams governance, integrations, and managed orchestration, but it is more structured and less transparent on pricing than open-source frameworks. If you want maximum customization and are comfortable building your own controls, a framework may fit better. If you want production-ready AI operations with less internal platform work, Kubiya is the stronger option.